An Overview on Audio, Signal, Speech, & Language Processing for COVID-19

Page created by Frank Freeman
 
CONTINUE READING
An Overview on Audio, Signal, Speech, & Language Processing for COVID-19
An Overview on Audio, Signal, Speech, & Language Processing for COVID-19
                                                                                  Gauri Deshpande1,2 , Björn W. Schuller2,3
                                                                                           1
                                                                                TCS Research India, India
                                         2
                                             Chair of Embedded Intelligence for Health Care and Wellbeing, University of Augsburg, Germany
                                                      3
                                                        GLAM – Group on Language, Audio, & Music, Imperial College London, UK
                                                                                   gauri1.d@tcs.com, schuller@ieee.org

                                                                   Abstract
                                         Recently, there has been an increased attention towards innovat-
                                         ing, enhancing, building, and deploying applications of speech
                                         signal processing for providing assistance and relief to human
arXiv:2005.08579v1 [cs.CY] 18 May 2020

                                         mankind from the Coronavirus (COVID-19) pandemic. Many
                                         ’AI with speech’ initiatives are taken to combat with the present
                                         situation and also to create a safe and secure environment for the
                                         future. This paper summarises all these efforts taken by the re-
                                         search community towards helping the individuals and the so-
                                         ciety in the fight against COVID-19 over the past 3-4 months
                                         using speech signal processing. We also summarise the deep
                                         techniques used in this direction to come up with capable so-
                                         lutions in a short span of time. This paper further gives an
                                         overview of the contributions from non-speech modalities that
                                         may complement or serve as inspiration for audio and speech             Figure 1: Speech and Audio Applications for COVID-19
                                         analysis. In addition, we discuss our observations with respect
                                         to solution usability, challenges, and the significant technology
                                         achievements.                                                        help in decelerating and eventually stopping the spread of the
                                         Index Terms: COVID-19, digital health, audio processing,             virus. Not only does identification and monitoring need digital
                                         computational paralinguistics, affective computing                   assistance, but also the post trauma phase would need it.
                                                                                                                   As per the WHO department of mental health and substance
                                                              1. Introduction                                 abuse, the current scenario of COVID-19 is susceptible to ele-
                                         As of today, there are more than 4 million confirmed cases of        vate the rates of stress or anxiety among individuals. Especially,
                                         coronavirus-induced COVID-19 infection cases in more than            lock-down and social distancing, quarantine and its after effects
                                         200 countries across the world. This led to increased need           in the society might have adverse effects on the levels of loneli-
                                         for screening, diagnostics, awareness, and post-care equipment       ness, depression, self-harm, or suicidal behaviour. A special at-
                                         and methods for a huge population. This not only demands for         tention is needed for the health care providers, having to face the
                                         reaching out to a larger population in shorter time, but also to     trauma directly and spending long working hours in such sce-
                                         come up with effective solutions at each stage of the pathologi-     narios. Both physical and mental health needs have increased
                                         cal, psychological, and behavioural problems. For the diagnosis      and require AI to provide faster and easy to access solutions.
                                         purpose, the world health organisation (WHO) 1 has described              The authors of [1] have already discussed about the oppor-
                                         the COVID-19 key symptoms as high temperature, coughing,             tunities, possible use cases, and the challenges that remain in
                                         and breathing difficulties, and an expanded list of symptoms         the space of speech and sound analysis by artificial intelligence
                                         that includes chills, repeated shaking with chills, muscle pain,     to help in this scenario of COVID-19. After this analysis was
                                         headache, sore throat, and a new loss of taste or smell. These       done, much more work has progressed in this direction which
                                         expanded list of symptoms are found to appear after 2-14 days        we are summarising in this paper. As depicted in Figure 1,
                                         of the onset of virus. However, the Coronavirus is so new to         we are giving an overview of the speech based applications for
                                         the mankind, that the unique symptoms of the virus and hu-           COVID-19 in this paper. Table 1 explains past speech related
                                         man immunity towards it are still found to vary from person to       work done to provide solutions for COVID-19 related health
                                         person. There are some cases found where a non-infected indi-        problems. This can be used as a ready reference by those who
                                         vidual with no symptoms can still be a carrier of the disease.       want to provide immediate solutions in this space. This table
                                              Due to such uncertainties, it is advisable not to completely    also describes the machine learning and deep learning technolo-
                                         depend on digital sensing to diagnose the virus, where many          gies used on the given data-sets to provide mentioned accuracy.
                                         physicians are also concerned about the false alarms that AI         For each speech or audio application, there is a vast space in the
                                         based diagnosing would generate. However, looking at the huge        literature. We have selected only a handful of them, considering
                                         number of infected individuals across the globe, detecting early     their relevance in the COVID-19 situation. Not everything that
                                         indications or screening will help in reducing this count. Con-      can be developed, can be used in this scenario considering other
                                         tinuous self-monitoring by the suspicious individuals can also       factors such as social distancing and personal and environmen-
                                                                                                              tal hygiene. Hence, the developments are required to be driven
                                             1 www.who.int                                                    by guidance from the clinicians and health care providers.
Table 1: Past speech and audio analysis related to COVID-19 health problems

 Application                         Ref.   Technology                        Dataset                            Accuracy
 Tuberculosis cough detec-           [2]    STFT, MFCC, MFB with              Google audio set extracts          .946 AUC
 tion                                       CNN                               from 1.8 million Youtube
                                                                              videos and Freesound audio
                                                                              database [3]
 Asthmatic detection                 [4]    INTERSPEECH             2013      Speech from 47 asthmatic           78 % Accuracy
                                            Computational         Paralin-    and 48 healthy controls
                                            guistics Challenge baseline
                                            acoustic features [5]
 Avoiding speech recording           [6]    PCA on audio spectro-             Acted cough from 17 pa-            True positive rate of 92 %
 while sensing cough and                    grams, FFT coefficients and       tients having cough due to         and false positive rate of
 hence providing privacy at-                RandomForest Classifier           common cold (n=8), asthma          0.5 %
 tribute                                                                      (n=3), allergies (n=1), and
                                                                              chronic cough (n=5)
 Obstructive sleep apnea             [7]    MFCCs with single layer           90 Male subjects’ speech           Cohen’s kappa coefficient of
 (OSA)     detection from                   neural network for breath-        and sleep quality measures         0.5 for breathing detection
 breathing sounds during                    ing detection and MFCCs,          using WatchPAT [8]                 and 0.54 for OSA detection
 speech                                     energy, pitch, kurtosis and
                                            ZCR with SVM for OSA
                                            classification
 Automatic speech breathing          [9]    Cepstrogram and Support           Speech recordings of 16            89 % F1 score and RMSE
 rate measurement                           Vector Machine with radial        participants of age group 21       of 4.5 breaths/min for the
                                            basis function                    years mean                         speech-breathing rate
 Breathing signal detection          [10]   Spectrogram with CNN and          20 healthy subjects’ speech        91.2 % sensitivity for breath
 for conversational speech                  RNN                               recorded using microphone          event detection and mean
                                                                              and breathing signal using         absolute error of 1.01
                                                                              two respiratory transducer         breaths per minute for
                                                                              belts                              breathing rate estimation
 Stress detection from speech        [11]   LLD and functional features       31 emergency call record-          Accuracy of 87.9 % with
 in situations such as car ac-              extracted using openSMILE         ings of the Integrated Res-        SVM and 87.5 % with CNN
 cident, domestic violence,                 [12] with k-NN, SVM, and          cue System of 112 emer-            in classifying stress from
 situations close to death,                 CNN classifiers                   gency line from Czech Re-          neutral speech
 and so on                                                                    public

2. Screening and Diagnosing for COVID-19                                     they had been tested positive. They would need these audio
                                                                             samples to build their machine learning algorithms for identify-
There exists a fine line between screening and diagnosing,
                                                                             ing COVID-19 cough. The Cambridge University’s web based
where screening gives an early indication of the presence of a
                                                                             platform asks the participants to also read a sentence so that
disease and diagnosing confirms the presence/absence of dis-
                                                                             their speech can also be recorded. Hence, their app consid-
ease. Screening is probabilistic, whereas diagnosis is binary in
                                                                             ers the sounds of cough, breathing and voice. Coughvid 4 is
nature. We will talk about different algorithms/applications us-
                                                                             another such app from EPFL (Ecole Polytechnique Fdrale de
ing audio processing for screening and diagnosis of COVID-19.
                                                                             Lausanne) to detect COVID-19 cough from other cough cate-
                                                                             gories such as normal cold and seasonal allergies. ’Breath for
2.1. Cough Detection                                                         Science’ 5 – a team of scientists from NYU –, have developed
Cough detection is not only identifying the cough sound and                  a web based portal to register the participants where they can
differentiate it from other sounds such as speech and laughter               enter similar details along with phone number. On pressing a
but also, to identify COVID-19 specific cough. This requires                 ’call me’ button, the participants receive a callback where they
to collect the COVID and non-COVID cough sound samples                       have to cough three times after the beep. With this, they plan
so as to develop an AI model that can differentiate between                  to create a dataset of cough sounds for the research purpose.
them on its own. For development of such a model, relevant                   Another web interface ’CoughAgainstCovid’ 6 for COVID-19
data needs to be collected. Cough audio samples can be col-                  cough sample collection is an initiative by Wadhwani AI group
lected using a simple smartphone mic, hence, major efforts are               in collaboration with Stanford University 7 .
seen in collecting it using smartphone mics such as done by the                   There are others who are ready with COVID-19 cough iden-
Carnegie Mellon University 2 and the Cambridge University 3 .                tifiers such as a smartphone app described in [13], which detects
Both the universities have provided platforms for the general                the COVID-19 cough.
population to upload their cough sounds along with some ad-
ditional information such as their age, gender, location, and if                4 https://coughvid.epfl.ch/
                                                                                5 https://www.breatheforscience.com/
   2 https://cvd.lti.cmu.edu/                                                   6 https://www.coughagainstcovid.org/
   3 https://www.covid-19-sounds.org/en/                                        7 https://www.stanford.edu/
Although these algorithms are attaining high accuracy lev-       2.4. Chest X-Ray
els of above 90 %, however, from the hygiene perspective it is
                                                                     Multiple efforts by several groups are put in the direction of de-
not advised to cough on open surface. As explained in [14], the
                                                                     veloping a classifier to detect COVID-19 symptoms using chest
infection is primarily transmitted through large droplets gener-
                                                                     X-Ray, such as that of in [19], [20], and many more. However,
ated during coughing and sneezing by symptomatic and also by
                                                                     as stated in [21], a multinational consensus is that Imaging is
asymptomatic people before onset of symptoms. These infected
                                                                     indicated only for patients with worsening respiratory status.
droplets can spread 1-2 m, deposit on surfaces and can remain
                                                                     Hence, it is advised for only those patients who have moderate-
viable for days in favourable atmospheric conditions but can
                                                                     severe clinical features and a high pre-test probability of dis-
be destroyed in a minute by common disinfectants. Hence, for
                                                                     ease. Hence, unlike Sections 2.1, 2.2, and 2.3, such measures
the collection procedures, it is important to remind the partici-
                                                                     are not suggestive for early identification and are preferred for
pants to cover the mouth and then only provide the cough sound
                                                                     diagnosis in a clinical setup.
samples. Otherwise, this may result in further spreading of the
                                                                          As concluded by [22], the chest X-ray findings in COVID-
disease. Also, after giving the samples, the smartphone surface
                                                                     19 patients were found to be peaked at 10-12 days from symp-
should be applied with an disinfectant.
                                                                     tom onset. Also, it is still required to visit a well-equipped
                                                                     clinical facility for such approaches. On a positive note, in the
2.2. Breathing Analysis                                              presence of mobile X-Ray machines, this approach can help in
As discussed before, shortness of breath is also one of the symp-    speeding up the diagnosis. The authors of [23] have found from
toms of the virus for which, the smartphone apps are designed        an experimental outcome that the chest X-Ray may be useful
to capture breathing patterns by recording their speech signal.      as a standard method for the rapid diagnosis of COVID-19 to
Breathing pattern detection has found applications in spectrom-      optimise the management of patients. However, CT is still lim-
etry to analyse lung functionalities as well. The authors of [15]    ited for identifying specific viruses and distinguishing between
have attempted to correlate high quality speech signals cap-         viruses.
tured using an Earthworks microphone M23 at 48 kHz with the
breathing signal captured using two NeXus respiratory induc-                               3. Monitoring
tance plethysmography belts over the ribcage and abdomen to
                                                                     The precautions for stopping the spread of the virus include so-
measure the changes in the cross-sectional area of the ribcage
                                                                     cial distancing, wearing a mask, and maintaining respiratory hy-
and abdomen at the sample rate of 2 kHz. They have achieved
                                                                     giene. Especially at public places, the concerned authorities can
a correlation of 0.42 to the actual breathing signal and a breath-
                                                                     monitor the population to confirm that the precautionary mea-
ing error rate of 5.6 % and sensitivity of 0.88 for breath event
                                                                     sures are adopted by them. This section describes such moni-
detection. In the Breathing Sub-challenge of Interspeech 2020
                                                                     toring tools developed for the COVID-19 scenario.
ComParE [16], the authors have achieved a baseline pearson’s
correlation of 0.507 on a development, and 0.731 on the test
dataset, respectively. They have used two piezoelectric respi-       3.1. Face Mask Detection from Voice
ratory belts for capturing breathing patterns. In addition to the    As described in Section 2.1, collecting cough samples without
speech signals, blow sound can also help in performing spirom-       covering mouth can lead to further spreading of the disease,
etry tests. The authors of [17] have developed a smartphone          mask wearing has to be mandated for donating a cough sam-
based automatic disease classification application built around      ple which requires an app to detect if the donor is wearing a
the spirometry tests.                                                mask or not. This year’s Interspeech 2020 ComParE challenge
                                                                     [16] features a mask detection sub-challenge, where the task is
2.3. Chat-Bots                                                       to recognise whether the speaker was recorded while wearing a
                                                                     facial mask or not. The results from this challenge will be use-
As the count of positive COVID-19 cases are increasing every         ful to develop a voice monitoring tool which detects the mask or
day, it brings up the need of automating the conversation a phys-    no-mask of a speaker. Also, these algorithms serve as a pre-step
iologist would have to understand the presence of symptoms in        for other speech processing algorithms such as speech recogni-
an individual. Microsoft and CDC have come up with a chatbot         tion, emotion recognition and cough detection.
named “Clara” for initial screening of COVID-19 patients by
asking them questions and capturing the responses. This uses         3.2. Cough and Breathing Analysis
speech recognition and speech synthesis technologies. The risk-
assessment test is designed based on advice from the WHO and         While at quarantine, the doctors need to monitor the cough his-
the Union Ministry of Health and Family Welfare India 8 . “Siri”     tory of patients, which can be done with a continuous cough
from Apple is also updated to answer the variations of the gen-      monitoring device. After we cross the crisis and organisa-
eral question, “Siri, do I have the coronavirus?” based on WHO       tions think of resuming the operations, continuous monitoring
guidelines. If a person shows severe symptoms, then it is ad-        of common spaces such as canteens and lobbies can be realised
vised by Siri to call 911. Similarly, Alexa is also updated with     to record the COVID specific coughs. One such monitoring ap-
answering COVID-19 screening question based on CDC guide-            plication is developed by the FluSense platform in [24]. It is
lines. Considering the trauma that the health care providers are     a surveillance tool to detect influenza like illness from hospital
going through, a web-based chat-bot named Symptoma devel-            waiting areas using cough sounds. Continuous monitoring and
oped in [18], is a significant step. It can differentiate 20 000     identification of abnormalities from the breathing rate has been
diseases with an accuracy of more than 90 % and can identify         done by [25] using image processing.
COVID-19 from other diseases having similar symptoms with
an accuracy of 96.32 %.                                              3.3. Mental Health – Emotion Detection
                                                                     The disease spread has equally affected the physical and men-
   8 https://www.mohfw.gov.in/                                       tal health of the individuals, which is primarily due to plenty
of mandatory precautionary measures such as social-distancing,                                4. Awareness
work from home and the quarantine procedures which usually
                                                                      Speech recognition and synthesis algorithms have been widely
takes at-least 15 days for the patients to be alone. Also, the
                                                                      used in the development of chat-bots to provide human like in-
health care providers owing to their hectic routines followed
                                                                      teractions. In this time of crisis, chat-bots are helping in spread-
by quarantine days are subject to undergo mental health issues.
                                                                      ing the valuable information about COVID-19 to end users.
To cater for the growing need of addressing this issue, not only
                                                                      Once an infected gets cured, they can help researchers with not
there is a high demand but the physical presence of the available
                                                                      only their experience but also with donating components such
psychologists is also missing. As found in [26], the COVID-19
                                                                      as plasma. CDC has been encouraging recovered people to do-
pandemic has generated unprecedented demand for tele-health
                                                                      nate their plasma for development of blood related therapies.
based clinical services.
                                                                      For collection of plasma, Microsoft has developed the chat-bot
     Among several initiatives taken against mental health is-        9
                                                                        which interacts with individuals to gather the required infor-
sues such as stress, anxiety, and depression, we are yet to see
                                                                      mation such as, the duration for which they are tested negative,
these emotions being analysed from the speech signal during the
                                                                      their age, weight, and also takes their pin-code to help them
COVID-19 period. This demands for the relevant data. Very re-
                                                                      know their nearest donation center.
cently, a study is conducted by the authors of [27] on the speech
signal of COVID-19 diagnosed patients. The behavioural pa-
rameters detected from speech includes, sleep quality, fatigue,                  5. Next Steps and Challenges
and anxiety considering corresponding self-reported measures          Since the work culture is moving more towards working-from-
as ground truth and have achieved an accuracy of 0.69. This           home, it will be required to detect some behavioural parame-
year’s ComParE challenge at Interspeech 2020 [16] has an              ters such as fatigue and sleepiness for the self monitoring of
Elderly Emotion Detection sub-challenge, where the speech             the working professionals. The behavioural parameters such as
captured from elderly people has to be classified into Low,           stress or anxiety need greater attention to be paid in these days.
Medium, and High classes of Valence and Arousal. It is found          A major challenge given the social distancing norms, is getting
that specific age groups such as that of elderly above 60 years       the relevant and accurate speech data for developing machine
are more prone to the infection, due to which this age group          learning models. Speech enabled chat-bots can play a signifi-
needs to follow the restrictions posed by the pandemic for a          cant role in this. There are other challenges as well from the
larger period in future as well. This shows that it will be crucial   design perspective of chat-bots. The authors of [32] have ex-
to understand the effects of this phase on their psychological        pressed both positive effect and drawbacks associated with us-
parameters such as emotions.                                          ing chatbots for sharing the latest information, encouraging de-
                                                                      sired health impacting behaviours, and reducing the psycholog-
3.4. Face Mask Detection from Image                                   ical damage caused by fear and isolation. This shows that the
One of the precaution measures while stepping outside home            design of chat-bots should be well thought of for using them,
suggested by the WHO to reduce the chance of getting infected         otherwise, they might have negative impact as well. An op-
or spreading COVID-19 is to wear a mask. Also, the mathemat-          timistic approach in these difficult times has been to work to-
ical analysis presented in [28] suggests that wearing a mask can      wards safe and secure environment for the post pandemic situa-
reduce community transmission and can help reduction in the           tion so that the society gain the trust and confidence back. This
peak death rate. Hence, similar to what has been outlined above,      shows the need of accurate and reliable screening and monitor-
it is important that the concerned authorities keep a check on        ing measures at public places.
people wearing masks or not at public places, especially where
the population is quite dense such as airports, railway stations,                             6. Conclusion
and hospitals.
     The authors of [29] have provided the Masked Face De-            There is a vast space of physical and mental wellness which
tection Dataset (MFDD), Real-world Masked Face Recognition            needs fast and usable solutions from the digital sensing space.
Dataset (RMFRD), and the Simulated Masked Face Recogni-               Most of the speech based research works for screening and
tion Dataset (SMFRD) for the detection of masked and un-              monitoring of COVID-19 are ready to be customised. Chat-
masked face using image processing techniques. They have              bots, although having the largest possibilities of use-cases, are
achieved 95 % accuracy in recognising masked faces. An Ap-            still in the need of context aware designs. As most of the first
ple store app developed by LeewayHertz can be integrated with         level symptoms are related to respiratory system, continuous
the existing setup of an Internet Protocol (IP) camera for getting    monitoring of the patients using audio signal processing based
alerts on detecting no-mask on a face. The system provides a          techniques can assist the clinicians or health care providers in
facility to add phone numbers for receiving the alerts and also       providing services while following social distancing norms. A
mechanism to see the face not wearing a mask for the admin.           multi-modal approach where all three modalities, audio and
                                                                      speech, text, and image together work towards creating a so-
3.5. Text Analysis                                                    lution framework will be likely the most promising avenue of
                                                                      future efforts.
The authors of [30] have used machine learning techniques to
categorise sentiments from the responses to Press Ganey patient
experience surveys. From this analysis, it is found that the pa-
                                                                                       7. Acknowledgements
tients have expressed negative comments about the cleanliness         We would like to thank all researchers, health supporters, and
and logistics and have given positive comments about their in-        others helping in this crisis. Our hearts are with those affected
teractions with clinicians. In [31], the twitter data with 5GCoro-    and their families and friends. We acknowledge funding from
navirus hashtag is analysed to understand the drivers of the 5G       the German BMWi by ZIM grant No. 16KN069402 (KIrun).
COVID-19 conspiracy theory and strategies to help in reducing
the spread of mis-information circulating in society.                    9 https://covig-19plasmaalliance.org/
8. References                                      [17] S. Trivedy, M. Goyal, P. R. Mohapatra, and A. Mukherjee, “De-
                                                                                sign and development of smartphone-enabled spirometer with a
 [1] B. W. Schuller, D. M. Schuller, K. Qian, J. Liu, H. Zheng, and             disease classification system using convolutional neural network,”
     X. Li, “Covid-19 and computer audition: An overview on what                IEEE Transactions on Instrumentation and Measurement, 2020.
     speech & sound analysis could contribute in the sars-cov-2 corona
     crisis,” arXiv preprint arXiv:2003.11117, 2020.                       [18] A. Martin, J. Nateqi, S. Gruarin, N. Munsch, I. Abdarahmane,
 [2] I. D. Miranda, A. H. Diacon, and T. R. Niesler, “A comparative             and B. Knapp, “An artificial intelligence-based first-line defence
     study of features for acoustic cough detection using deep architec-        against covid-19: digitally screening citizens for risks via a chat-
     tures,” in 2019 41st Annual International Conference of the IEEE           bot,” bioRxiv, 2020.
     Engineering in Medicine and Biology Society (EMBC). IEEE,             [19] P. Afshar, S. Heidarian, F. Naderkhani, A. Oikonomou, K. N. Pla-
     2019, pp. 2601–2605.                                                       taniotis, and A. Mohammadi, “Covid-caps: A capsule network-
 [3] E. van Miltenburg, B. Timmermans, and L. Aroyo, “The vu sound              based framework for identification of covid-19 cases from x-ray
     corpus: Adding more fine-grained annotations to the freesound              images,” arXiv preprint arXiv:2004.02696, 2020.
     database,” in Proceedings of the Tenth International Conference       [20] I. D. Apostolopoulos and T. A. Mpesiana, “Covid-19: automatic
     on Language Resources and Evaluation (LREC’16), 2016, pp.                  detection from x-ray images utilizing transfer learning with con-
     2124–2130.                                                                 volutional neural networks,” Physical and Engineering Sciences
 [4] S. Yadav, M. Keerthana, D. Gope, P. K. Ghosh et al., “Analysis             in Medicine, p. 1, 2020.
     of acoustic features for speech sound based classification of asth-   [21] G. D. Rubin, C. J. Ryerson, L. B. Haramati, N. Sverzellati, J. P.
     matic and healthy subjects,” in ICASSP 2020-2020 IEEE Inter-               Kanne, S. Raoof, N. W. Schluger, A. Volpi, J.-J. Yim, I. B. Martin
     national Conference on Acoustics, Speech and Signal Processing             et al., “The role of chest imaging in patient management during
     (ICASSP). IEEE, 2020, pp. 6789–6793.                                       the covid-19 pandemic: a multinational consensus statement from
 [5] B. Schuller, S. Steidl, A. Batliner, A. Vinciarelli, K. Scherer,           the fleischner society,” Chest, 2020.
     F. Ringeval, M. Chetouani, F. Weninger, F. Eyben, E. Marchi
                                                                           [22] H. Y. F. Wong, H. Y. S. Lam, A. H.-T. Fong, S. T. Leung, T. W.-
     et al., “The interspeech 2013 computational paralinguistics chal-
                                                                                Y. Chin, C. S. Y. Lo, M. M.-S. Lui, J. C. Y. Lee, K. W.-H.
     lenge: Social signals, conflict, emotion, autism,” in Proceedings
                                                                                Chiu, T. Chung et al., “Frequency and distribution of chest ra-
     INTERSPEECH 2013, 14th Annual Conference of the Interna-
                                                                                diographic findings in covid-19 positive patients,” Radiology, p.
     tional Speech Communication Association, Lyon, France, 2013.
                                                                                201160, 2020.
 [6] E. C. Larson, T. Lee, S. Liu, M. Rosenfeld, and S. N. Patel, “Ac-
     curate and privacy preserving cough sensing using a low-cost mi-      [23] Y. Li and L. Xia, “Coronavirus disease 2019 (covid-19): role
     crophone,” in Proceedings of the 13th international conference on          of chest ct in diagnosis and management,” American Journal of
     Ubiquitous computing, 2011, pp. 375–384.                                   Roentgenology, pp. 1–7, 2020.
 [7] R. M. Simply, E. Dafna, and Y. Zigel, “Obstructive sleep ap-          [24] F. Al Hossain, A. A. Lover, G. A. Corey, N. G. Reich, and T. Rah-
     nea (osa) classification using analysis of breathing sounds during         man, “Flusense: A contactless syndromic surveillance platform
     speech,” in 2018 26th European Signal Processing Conference                for influenza-like illness in hospital waiting areas,” Proceedings of
     (EUSIPCO). IEEE, 2018, pp. 1132–1136.                                      the ACM on Interactive, Mobile, Wearable and Ubiquitous Tech-
                                                                                nologies, vol. 4, no. 1, pp. 1–28, 2020.
 [8] Y. J. Gan, L. Lim, and Y. K. Chong, “Validation study of watchpat
     200 for diagnosis of osa in an asian cohort,” European Archives of    [25] Y. Wang, M. Hu, Q. Li, X.-P. Zhang, G. Zhai, and N. Yao, “Ab-
     Oto-Rhino-Laryngology, vol. 274, no. 3, pp. 1741–1745, 2017.               normal respiratory patterns classifier may contribute to large-scale
 [9] A. Routray et al., “Automatic measurement of speech breathing              screening of people infected with covid-19 in an accurate and un-
     rate,” in 2019 27th European Signal Processing Conference (EU-             obtrusive manner,” arXiv preprint arXiv:2002.05534, 2020.
     SIPCO). IEEE, 2019, pp. 1–5.                                          [26] P. D. Patel, J. Cobb, D. Wright, R. Turer, T. Jordan, A. Humphrey,
[10] V. S. Nallanthighal, A. Härmä, and H. Strik, “Deep sensing of            A. L. Kepner, G. Smith, and S. T. Rosenbloom, “Rapid develop-
     breathing signal during conversational speech,” in Proceedings of          ment of telehealth capabilities within pediatric patient portal in-
     the 20th Annual Conference of the International Speech Commu-              frastructure for covid-19 care: Barriers, solutions, results,” Jour-
     nication Association, Interspeech. Graz, Austria:[Sn], 2019, pp.           nal of the American Medical Informatics Association, 2020.
     4110–4114.                                                            [27] J. Han, K. Qian, M. Song, Z. Yang, Z. Ren, S. Liu, J. Liu,
[11] P. Partila, J. Tovarek, J. Rozhon, and J. Jalowiczor, “Human stress        H. Zheng, W. Ji, T. Koike et al., “An early study on intelligent
     detection from the speech in danger situation,” in Mobile Mul-             analysis of speech under covid-19: Severity, sleep quality, fatigue,
     timedia/Image Processing, Security, and Applications 2019, vol.            and anxiety,” arXiv preprint arXiv:2005.00096, 2020.
     10993. International Society for Optics and Photonics, 2019, p.       [28] S. E. Eikenberry, M. Mancuso, E. Iboi, T. Phan, K. Eikenberry,
     109930U.                                                                   Y. Kuang, E. Kostelich, and A. B. Gumel, “To mask or not to
[12] F. Eyben, M. Wöllmer, and B. Schuller, “Opensmile: the mu-                mask: Modeling the potential for face mask use by the general
     nich versatile and fast open-source audio feature extractor,” in           public to curtail the covid-19 pandemic,” Infectious Disease Mod-
     Proceedings of the 18th ACM international conference on Mul-               elling, 2020.
     timedia, 2010, pp. 1459–1462.
                                                                           [29] Z. Wang, G. Wang, B. Huang, Z. Xiong, Q. Hong, H. Wu, P. Yi,
[13] A. Imran, I. Posokhova, H. N. Qureshi, U. Masood, S. Riaz,                 K. Jiang, N. Wang, Y. Pei et al., “Masked face recognition dataset
     K. Ali, C. N. John, and M. Nabeel, “Ai4covid-19: Ai enabled pre-           and application,” arXiv preprint arXiv:2003.09093, 2020.
     liminary diagnosis for covid-19 from cough samples via an app,”
     arXiv preprint arXiv:2004.01275, 2020.                                [30] S. Guney, C. Daniels, and Z. Childers, “Using ai to understand
                                                                                the patient voice during the covid-19 pandemic,” NEJM Catalyst
[14] T. Singhal, “A review of coronavirus disease-2019 (covid-19),”             Innovations in Care Delivery, vol. 1, no. 2, 2020.
     The Indian Journal of Pediatrics, pp. 1–6, 2020.
                                                                           [31] W. Ahmed, J. Vidal-Alaball, J. Downing, and F. L. Seguı́, “Dan-
[15] V. S. Nallanthighal, A. Härmä, and H. Strik, “Speech breathing
                                                                                gerous messages or satire? analysing the conspiracy theory link-
     estimation using deep learning methods,” in ICASSP 2020-2020
                                                                                ing 5g to covid-19 through social network analysis,” J. Med Inter-
     IEEE International Conference on Acoustics, Speech and Signal
                                                                                net Res, 2020.
     Processing (ICASSP). IEEE, 2020, pp. 1140–1144.
[16] B. W. Schuller, A. Batliner, C. Bergler, E.-M. Messner, A. Hamil-     [32] A. S. Miner, L. Laranjo, and A. B. Kocaballi, “Chatbots in the
     ton, S. Amiriparian, A. Baird, G. Rizos, M. Schmitt, L. Stappen            fight against the covid-19 pandemic,” npj Digital Medicine, vol. 3,
     et al., “The interspeech 2020 computational paralinguistics chal-          no. 1, pp. 1–4, 2020.
     lenge: Elderly emotion, breathing & masks,” Proceedings INTER-
     SPEECH. Shanghai, China: ISCA, 2020.
You can also read